A compact optical neuron based on multi-operand microring resonators (Conference Presentation)

Integrated photonics has emerged as a promising candidate for next-generation AI hardware accelerators that can overcome the bottlenecks of their electrical counterparts. However, current photonic-electronic optical neural networks (ONNs) face several practical challenges, including a large footprint, the necessity of electronic peripherals, and inefficient electronic-optical (E-O) interfaces. These limitations significantly restrict their application in complex AI tasks due to their limited scalability. In this work, we propose an optical neuron characterized by a compact footprint, high scalability, and built-in nonlinearity using multi-operand microring resonators (MOMRRs). Based on on-chip validation, we experimentally demonstrate the effectiveness of MOMRRs, which enhances the compactness, hardware efficiency, and scalability of ONNs at the device level.